An approach to anomalous diffusion in the n-dimensional space generated by a self-similar Laplacian

An approach to anomalous diffusion in the n-dimensional
space generated by a self-similar Laplacian

Thomas M. Michelitsch 1 * Gerard A. Maugin 1

Andrzej F. Nowakowski 2 , Franck C. G. A. Nicolleau 2

Mujibur Rahman 3

1 Universite Pierre et Marie Curie, Paris 6
Institut Jean le Rond d'Alembert
CNRS UMR 7190
France

2 Sheffield Fluid Mechanics Group
Department of Mechanical Engineering
University of Sheffield
United Kingdom

GE Energy, Greenville, SC 29615

^: usa

Submitted manuscript

michel-diff.tex

August 1, 2012

* Corresponding author, e-mail : michel@lmm.jussieu.fr

1

Abstract We analyze a quasi-continuous linear chain with self-similar distribution of harmonic
interparticle springs as introduced for one dimension in [§]. We define a continuum limit for one di-
mension and generalize it to n = 1,2,3,.. dimensions of the physical space. In this way we obtain
a Hamiltonian functional in the n = 1,2,3, .. dimensional physical space: Application of Hamilton's
(variational) principle defines then a self-similar and as consequence non-local Laplacian operator for
the n- dimensional space. By employing this Laplacian we establish a Fokker Planck diffusion equation:
We show that this Laplacian generates spatially isotropic Levi stable distributions which correspond to
Levi flights in n- dimensions. In the limit of large scaled times ~ t/r a » 1 the obtained distributions
exhibit an algebraic decay ~ — > independent from the initial distribution and spacepoint. This
universal scaling depends only on the ratio n/a of the dimension n of the physical space and the Levi
parameter a.

Keywords. Fokker Planck equation, anomalous diffusion, Levi flights, Levi (stable) distributions,
self-similarity, scaling laws, fractional operator, non-locality

1 Introduction

It has been recognized that complex processes exist where the role of fluctuations is by far underes-
timated and which cannot be described by Gaussian statistics. These random particle motions have
much more erratic characteristics as the "benign" Brownian motions which can be described by Gaus-
sian statistics and normal distributions. This is true for instance for the stock market, where Gaussian
statistics fails largely by underestimating fluctuations (risks!). One of the first researchers to recognize
this was indeed Mandelbrot [T] |2]. Among the stochastic motions that are characterized by infinite
mean fluctuations are those with Levi-distributed scale free jump distributions. Such motions, known
in the literature as Levi flights are the source of anomalous diffusion pTJ|. Levi flights are widely found
in nature for instance in the dynamics of bumblebees [7J . There is a huge number of models of Levi
flights by fractional kinetic equations of diffusion and a vast literature devoted to this subject. An
excellent overview is provided in [17] (see also the references therein). However, the subject of the
present paper is not the investigation of the properties of Levi flights themselves which are well studied,
e.g. [5J El H7J- In most of these models a diffusion problem is ad hoc defined such that Levi stable
distributions solve the fractional problem. However, a physical interpretation or justification of these
fractional operators is in many cases not given [5], [6]. The goal of the present paper is to establish
a simple and physically well based model where the Laplacian used in the Fokker Planck diffusion
equation is rigorously deduced from Hamilton's principle.

This paper is organized as follows: We deduce a self-similar Laplacian as a result of a continuum
limit of the linear chain with a self-similar distribution of interparticle springs in one dimension.
This continuum limit results in a Laplacian in the form of a self-adjoint combination of fractional
operators. Then we generalize this Laplacian to the n-dimensional physical space (n = 1,2,3). The
generalization to higher dimensions than one is performed such that the Laplacian maintains spatially
isotropic symmetry. We deduce the dispersion relation and the density of normal modes (oscillator
density) both obeying characterstic scaling laws. By employing this Laplacian we formulate a diffusion
equation (Fokker Planck equation) which generates Levi stable distributions. It does not matter for
the model whether we conceive this distribution as particle density function or probability density
function. In the first case we consider an ensemble of diffusing particles and in the other case the
probability distribution of one propagating particle. For the latter case the stochastic particle motion
which describes our model are Levi flights. Despite we refer more to the "diffusion picture", the
interpretation of our model can be fully transferred into the "stochastic picture" of a single randomly
walking Levi flyer particle.

We derive the time evolution operator in its space-time representation. We focus especially on the
asymptotic behavior at large (scaled) times which is found to be a spatially uniform algebraic decay

2

approaching zero density. We demonstrate that this characteristic spatially homogeneous behavior of
the asymptotic density indicates the approach of the system to a quasi-equal distribution approaching
maximum (infinite) entropy due to the complete uncertainty of the location of a particle independent
from the initial distribution.

To deduce the self-similar Laplacian operator we consider a "material system" with a spatially
homogeneous constant mass distribution and spatially harmonic self-similar interparticle interactions.
As demonstrated in [9] such a material system exhibits a regime with a fractal dispersion relation that
appears in a Weierstrass-Mandelbrot function which is exactly self-similar. Interparticle interactions
which are exactly self-similar in the below defined sense require an infinite physical space. In the
continuum limit of this model the fractal properties of the dispersion relation disappear and the
dispersion relation assumes the form of a smooth power function [§J[TTJ[T2]. ^ e con ti nuum limit of this
model opens a general way to deduce continuous field theories in media with self-similar interparticle
interactions leading to self-similar constitutive laws. In a recent paper [TT] this theory is elaborated
for one dimension. In the present paper we generalize this model to n = 1,2,3,.. dimensions of
the physical space. For the sake of simplicity we refer to the medium with self-similar interactions
just as "self-similar medium" or "self-similar continuum". We emphasize that the self-similarity of
interparticle interactions is the inevitable source of non-locality in the deduced Laplacian operator. In
contrast to this, a linear chain which contains only harmonic next neighbor interparticle interactions
yields in the continuum limit the traditional spatially local Laplacian. A discrete material system
leading to a fractal dispersion relations due to non-local interparticle interactions has probably first
been considered by Tarasov [13] (see also the numerous references therein).

2 Preliminary remarks

It is an interesting question how physical phenomena change when the interparticle interactions become
self-similar. For instance when we consider the Poisson equation

with a self-similar Laplacian operator A se ifsim which is to be specified. To this end we should elaborate
first of all the notion of "self-similarity" employed in this paper. The notion of " self- similarity" which
is employed in this paper as well as in [TTJ 02] corresponds to the notion " self- similarity at a
point 1 which is commonly used in the mathematical literature [13]. Generally, an object is exactly
self-similar in the strict sense if it can be decomposed into parts which are exact rescaled copies of
the entire object. In contrast is the notion "self-similarity at a point" where "point" means here a
fixed-point of the scaling operation: An object which is self-similar at a point contains a single part
which is a re-scaled copy of the entire object and so forth over an infinity of scales [T5] .

According to this notion of self-similarity we call a function A(h) self-similar with respect to h, i.e.
self-similar at point h = 0, when the relation

is fulfilled for a prescribed scaling factor N > 1 and for any h > 0. We assume real valued scaling
exponents 5 G K. If a function A(h) fulfills (j2]), it follows that fl2]) remains true if we replace N — > N s
with (s G Zo denotes positive and negative integers including zero). In other words: if A(h) is self-
similar with respect to h in the sense of relation (j2J), then there exists a iV > 1 such that only the
discrete set of rescaling operations h! = hN' with N' = N s with positive and negative integers s G Z
including the zero, satisfy the self-similarity condition namely A(hN') = N' s A(hj^. We observe
further by putting h = N p+X with p G Zo and < \ < 1 denoting the non- integer part, and by using
the property of self-similarity that (N p = hN~ x )

^dn the continuum limit to be discussed below, this set {hN s } becomes a continuous one.

A(Nh) = N 5 A(h)

(2)

3

A(h = N p N x ) = N pS A(N x ) = h s N~ Sx A(N x )

(3)

From this relation we observe that all values of A are uniquely determined by its values within
1 < iV x < N (as < x < 1) [10]. Further we see that A(h) scales as h s , especially when h — > 0,
however a unique limit h — > of A(h) does not exist because of its dependence on x of h(x,p) = N p+X .
Let us assume that a constant C > exists such that < N~ Sx A(N x ) < C Wh > then function A(h)
fulfills the inequality

< A{h) < Ch 5 (4)

For < 5 < 1 relation (jlj) is the Hoelder condition, e.g. [3]. The function A(h) is then a Hoelderian
function (Hoelder continuous function) being continuous but non-differentiable for < 5 < 1 at h = 0.
Hoelderian functions include a wide range of fractal and erratic functions [3]. For 5 > 1 the function
A(h) is differentiate at h = 0. Self-similar functions can hence be fractal or non-fractal functions.

Generally a self-similar function which fulfills ([2]) for a prescribed N can be written in the form

oo

A(h) = J2 N~ 5s f(N s h) (5)

s=— oo

which converges for sufficiently good functions / [9]. Without any loss of generality we can restrict
ourselves to iV > 1. The simplest self-similar functions of this type are power-functions h s . They
constitute the continuum limit (below equation Q) of (jSJ).

3 The self-similar elastic continuum
3.1 One-dimensional case

In this section we evoke a self-similar Laplacian from a simple linear chain model which was introduced
in [9] and its continuum limit we introduced in recent papers [TTJ Q2] for the one- dimensional infinite
space. We consider this quasi-continuous chain with self-similar harmonic springs with the Hamiltonian
functional [9]

H = - (u 2 {x,t) + V{x,t,h)) dx (6)

Z J— oo ^ '

where x denotes the space- and t the time coordinates. Each spacepoint x denotes also a mass point
and the mass density of the system we consider is constant (spatially homogeneous) and we put the
mass density equal to 1. \V(x, t, h) indicates the elastic energy density with

1 oo

V{x,t,h) = ~ J2 N- 5s [{u(x,t)-u(x + hN s ,t)} 2 + {u(x,t)-u(x-hN s ,t)} 2 ] (7)

! = — OO

which converges in the range < 5 < 2 and where we assume h > and N being a prescribed scaling
factor. Without loss of generality we can restrict ourselves to > 1 (N G K). The additional factor
1/2 in the elastic energy fl7]) compensates double counting of the springs when integrating in flS]). It is
important to note that the elastic energy density does not have any characteristic interaction length
scale. The variable h characterizes self-similarity of the elastic energy density, but does not have the
physical meaning of a characteristic length. Unlike in the case of a linear chain with only next neighbor
interactions, the limit h — > would not localize the interparticle interactions. In and (j7|) u and
ii = -j^u stand for the displacement field and the velocity field, respectively. ([7j) has the property of
being self-similar with respect to h at point h = 0, namely

4

V{x,t,Nh) = N s V(x,t,h) (8)

As a starting point for the approach to be developed we evoke the continuum limit of (|7|) and the
resulting equation of motion. For 0<5<l(J7])isa Hoelder continuous function being for < 5 < 1
non-differentiable at h = 0. If we prescribe in (JZj) a periodic field u(x) then the elastic energy density
is in < 5 < 1 a fractal function (Weierstrass-Mandelbrot fractal function). We define the continuum
limit asiV = l + £(0<C<<l)soT = hN s becomes a continuous variable and we can write a
self-similar function A(h) which fulfills a self-similarity condition (JS} asymptotically as [9J

A(h) = £ N~ Ss f(N°h) ~ - / ^dr (9)

s=-oo S> J0 T

having the form of a power function A(h) = const h s . Both the discrete as well as the continuous
representation of converge for sufficiently good functions (see details in [9]). From follows that
in that continuum limit we can write (j7|) as a functional of the displacement field u(x,t) in the form

h s r<* {{u{x, t) - u{x + t t)f + (m(x, t) - u{x - r, t)) 2 }
V(x, t, /i) = — ^ dr (10)

which exists as in the discrete case (JZ]) in the band < 5 < 2. Application of Hamilton's principle
leads then to the definition of the Laplacian of our system which is then determined by the functional
derivative of the elastic energy V with respect to the field u, namely [18]

A(W)«=:-^, V = lJ Vdx ( U )
The equation of motion (self-similar wave equation) has then the form [9]

d 2

— u(x, t) = A [SM u(x, t) (12)

pip together with ffTOl) defines the self-similar Laplacian in its continuous representation of the
one-dimensional medium. Relation ( llip defines the Laplacian for both cases, in the the discrete case
of (jTJ) as well as for the continuous case ffTUj) . In both cases the Laplacian is due to its construction
self-similar with respect to h. The self-similar Laplacian is necessarily a non-local and self-adjoint
negative semi-definite^|, spatially isotropic operator: In the continuous case it is obtained as

h 5 r°° (u(x - r) + u(x + r) — 2u{x))

~C Jo T

existing for < 5 < 2. It might be sometimes convenient to rewrite (TIB"]) in the equivalent form

A(<5acW) = — /„ Zi+s dr ( 13 )

h 5 d r°° (u(x + t) — u(x — t)) , .

A ««>"M = ^/ u V ^ dT ' 0<i<2 (14)

where the range of existence of these relations is < 5 < 2. In the entire analysis of this paper we put
for any complex number z = \z\e tip the principal value —n < ip = Arg(z) < 7r for its argument (p. In
the further analysis we will need the T-function (faculty-function) T(z) which is defined by [16J

T(a + 1) =: a\ = / e" r r a dr, Re(a) > -1 (15)
Jo

Uniform translations are eigenmodes to eigenvalue zero.

5

The condition Re(a) > — 1 is required for integral (|T5|) to exist. Re(Z) denotes the real- and Im(Z)
the imaginary part of a complex number Z. Using (I14p the equation of motion (j!2p takes the form

d 2 d

-^u(x,t) = —a(x,t) (16)

where ^ indicates the traditional partial derivative with respect to x and a(x,t) denote the stress
having the form [12]

h 5 r°° (u(x + t) — u(x — t)) , „ i .

o{x) = -l U ^ dr , <5<2 (17)

where the integration constant turns out to be zero when integrating the right hand side of (|14j) with
respect to x. For further comparision it will be convenient to represent (|17|) in the equivalent form

h 5 r°° (u(x + t) — u(x — t)) , . . .

ffW = 2a/oo rU ^ ^ 0< " <2 (18)

where sgn(r) = m maintains the integrand to be an even function with respect to r.

From this follows that self-similarity of the interparticle interactions leads inevitably to non-local
field theories. In the elastic framework the material (moduli) functions are convolution kernels. In
contrast to the "classical" non-local elasticity theory as outlined by Eringern [15] the self-similar case
is characterized by long-range power law kernels decaying critically slowly [TT| [T2] .

4 The self-similar Laplacian of the n- dimensional space

We introduce a generalization of the self-similar Laplacian f[T5|) to the n- dimensional space where
n = 1,2,3. We define this Laplacian by its action on a scalar field variable w(x). Then we can
generalize the one-dimensional case ( TIB"]) to n = 1, 2, 3 dimensions as

A (n , a)U (x) = - J ^ 1 ^d"r, < 5 - (n - 1) < 2 (19)

This integral is performed over the entire n-dimensional physical space IR n . The prefactor 2 _1 has been
chosen due to the fact that in (fT3"|) a prefactor 2 _1 has to be added to the integrand if we integrate over
the entire physical space R . (T3~9|) exists for sufficiently smooth fields w(x) converging in the interval
n — 1 < 5 < n + 1. In some cases also the equivalent representation

A M) u(x) = - / 1 | _ d"r, < S - (n - 1) < 2 (20)

/• Hr)-« (x)} JB-
C y |r-x|

might be convenient. For the analysis the following representation of fTl9|) in terms of a divergence of
a vector field will be useful

A (rM5)M (x) = V x • D(x) (21)

where (V x )j = gf- denotes the (traditional) gradient operator. The vector field D is determined (up
to an unimportant rotational gauge vector field b with V ■ b = 0) by

D x

(5-(n- 1)) 2C

y" ^ {w(x + r) -u(x-r)} d n r, 0<<J-(n-l)<2 (22)

which recovers for n — 1 expression (|18|) . This vector field exists in the same interval n — 1 < 5 < n + 1
as ( TT9"j) . The deduction of (|22p is performed in the appendix 18.11 by using the Gaussian theorem.

6

Further useful equivalent representations of (j22j) are given in the appendix. We note that u is a
scalar field and the integrand of (122 p is an even function with respect to integration variable r. For
n = 1 (T2"2"|) corresponds to the stresses ( TTS|) . If we conceive equation |T| as a Poisson equation in
an electrodynamic context then the vector field ([22]) can be conceived as the "dielectric displacement
field". Then ([1]) with f[2~T]) defines the self-similar Gauss-law (charge conservation). Starting from
this we can set up a theory of self-similar fields and set up "self-similar" Maxwell equations. In the
appendix 18.11 we have deduced a useful scalar "potential" (eq. f ]8"3"]) ). However, we will elaborate the
subject of a "self-similar" electrodynamic field theory in a sequel paper.

4.1 Dispersion relation and density of normal modes

In view of the translational symmetry of the Laplacian, we observe that plane- waves 0fc(r) = e* kr are
eigenfunctions of the Laplacian f lT9|) where its negative eigenvalues constitute the dispersion relation
u 2 (k) which is obtained by the relation

A n ,^(r) = -< 5 (A;)0 fc (r) (23)

h s (e ik - r + e - ik r - 2 )
= / USX l ^ ( 24 )

2( J r 8+1

We observe that the dispersion relation fulfills the scaling property

col s (k) = A ntS k 5 - n+1 , 0<5-n + l<2 (25)

depending only on k = |k| reflecting isotropy of the Laplacian and A n j = o^^/c = 1) * s defined by
([21]). We observe from (|2~21) that the coefficient A Uj s > is strictly positive indicating "elastic stability".
A further evaluation of the coefficient A n> g is given in the appendix 18. 2[ The following observation
is noteworthy: In dispersion relation [2B\) appears always a positive exponent < <5 — n + 1 < 2
being within the interval (0, 2) whatever the dimension n of the physical space. From this follows the
important property

-+ 0) = (26)

for any dimension n as a necessary consequence of the translational invariance of the self-similar
Laplacian. Translational invariance requires that the k = mode (uniform translation of the entire
"material system") must give a zero contribution to the elastic energy and hence corresponds to
eigenvalue zero.

It is now straight-forward to obtain the density of normal modes which we denote by T>(oj). This
quantity is defined such that V(u)du counts the number of normal modes (per n-dimensional unit-
volume) with frequencies in the interval [u,w + da;]. We obtain this quantity from the dispersion
relation by (e.g. [H] or any textbook of theoretical physics)

V(co)doo = J—Onil^-'dk (27)
where we put a = S — (n — l) where k = k(cu) = ^ is the inverse dispersion relation and O n (l) = Ir^v

A a 1 I 2 I

denotes the surface of the unit-sphere. Then (j27j) yields the scaling law

= — u-- 1 28

K>T{*)aA« 5

7

where the prefactor is always positive and the exponent ^ — 1 > n — 1 > (0 < a < 2) whatever the
dimension n of the physical space. Hence we have for n = 1 an exponent | — 1>0(0<5<2); for
n = 2 an exponent -j—j- — 1 > 1 (1 < S < 3); and for n = 3 an exponent ~~ 1 > 2 (2 < <5 < 4).
The above power law ([28]) confirms our conjecture raised in [9], namely that the property

V(u ->• 0) = (29)

holds generally in self-similar media where the density of normal modes obeys a power law with
positive exponent for any dimension n of the physical space. For n — 1 the expression obtained in [9]
is recovered by fj29|) with the exponent being | — 1 > (0 < £ < 2) and yields

V x {u) = 2 t w^ 1 (30)

In contrast to the self-similar Laplacian of our model the traditional Laplacian would give a dis-
persion relation u 2 (k) = k 2 and correspondingly the density of normal modes would be in the n-
dimensional space

^traditional^) = ^n)" ^ 1 = ^fp^n^" 1 0*1)

scaling as ~ u n ~ l and where (1251) would assume (13T!) when we put there the "forbidden" value a = 2
and A = 1. We emphasize that in a space of dimension n the exponent due to the self-similar (non-
local) Laplacian is always greater than the exponent due to the traditional (localized) Laplacian

In

l>n-l, 0<a = 5-(n-l)<2 (32)

a

It is quite remarkable that the self-similar Laplacian gives rise to unusual physical phenomena being
qualitatively different from those found with a traditional Laplacian.

Whereas diffusion problems formulated with a traditional Laplacian describe traditional Gaussian
statistics with finite variances, diffusion problems formulated with Laplacian ffl~9]) describe random
motions allowing long-range scale-free distributed jumps and yielding infinite variances referred to as
Levi flights. We devote the next section to this problem.

5 Anomalous diffusion problem in n dimensions - Levi flights

We consider an ensemble of particles of density p(r, t) where p(r, t)d n r denotes the number fraction of
particles being located at a time t in the volume element d"r which is attached to spacepoint r. We
refer to this picture as the "diffusion picture". The other picture of this model is a single randomly
walking particle where p(r,t)d™r then means in this "stochastic picture" the probability to find the
particle at time t > in the volume element d n r which is attached to spacepoint r. In what follows
we describe a model of the space-time evaluation of the density p where both pictures, the diffusion
picture and the stochastic picture are possible physical intepretations. In the stochastic picture the
present model discribes a random walking particle which is walking continuously in time where the
jump distance of the particle in any infinitesimal time interval St is distributed according to a power
law and isotropic in space. For such stochastic motions Mandelbrot coined the term Levi flights. We
will see that our above introduced self-similar Laplacian operator is the source of exactly this type of
stochastic motion in the n = 1, 2, 3, ..-dimensional space.

We consider the problem for t > 0. We are especially interested in the characteristic asymptotic
behavior for large times t.

8

The density to be analyzed is normalized for all times t > according to

|p(r,t)d"r = l (33)

We then define the diffusion problem by the following diffusion equation (in the stochastic picture
Fokker Planck equation)

^p(r,t) = -£ n , sP (r,t) (34)

where — £ n ,<5 = ^n,5 denotes the Laplacian (1191) . We further assume a prescribed initial distribution

p(r,t = 0)=p o (r) (35)

which is also normalized according to (I33p . We conceive the positive semi-definite operator # as
the diffusion generator where its eigenvalue spectrum is just the dispersion relation (125]) with the
diffusional eigenmodes 0/c(r) = e ikr . The density p(r,t) writes in terms of eigenmodes as Fourier
transformation

p(r,t) = ^/p(k,t)e' k - r d"k (36)
the Fourier amplitude p(k, t) fulfills the evolution equation

^p(k,t) = -ul 5 (k)p(k,t) (37)

showing an exponential decay in time (A = A n ^ > 0)

p(k,t) = e- tAka p Q (k) (38)

where we have put a = <5 — (n — 1) and po(k) indicating the Fourier transform of the initial distribution
Po(r). Unlike in the case of Gaussian diffusion, the non- locality of Laplacian (T191) indicates that non-
local particle jumps are admitted which are scale free distributed. We will come back to this important
property more closely below.

We can also write the solution of (EH) in the form

p(r,t) = e- tc ^p (r) (39)
which we will evaluate next. We observe that the normalization of (1391) is maintained at all times t >

J p(r,t)d n r = e~ tCn - 6 J p (r)d"r = le~ Cn - 6t l = 1 (40)

since Po(r) is normalized with C n ^ = so that all powers higher than m = in the exponential
series of the time evolution operator e~ Cn - st l applied on a constant yield vanishing contributions. This
indicates that the total particle number is a conserved quantity. It is further illuminating to consider
the diffusion processes more closely: To this end we consider how the density p(r, t + d~t) evolves from
the density p(r, t) where 6t is an infinitesimal small time interval

p(r,t + 6t) = e- 5tCn - s p(r,t) (41)
where we can put e~ 5tCn - s ~ 1 — 5tC n ^ and so

p(r, t + 5t) = p(r, t) - 5t£ niS p(r, t) (42)

9

The negative Laplacian C n> $ is the generator, generating the infinitesimal transformation of the density
from t to t + St.

Since p(r,t)d n r denotes the particle number fraction being at time t in volume element d n r which
is attached to the spacepoint r, we can conceive the quantity

p(r,t + St) - p(r,t) 9

si dt p ('> = ~ £ ".*/ J ( r '*) ( 43 )

as the net balance of the particle number(fraction) departing and arriving in volume element d n r
during the time interval St. In other words (I43p measures the number of particles jumping into the
volume element minus the number of particles jumping out of the volume element.

The local net balance is due to the non-locality of the generating operator £ n ^p(r, t) depending on
all values of p at time t in the entire physical space R n and not (as in the case of Gaussian diffusion)
only from the p-values in the local neighborhood. We can express this local balance described by the
diffusion equation fl34"|) in terms of a continuity equation

j^(x,f) = -V..J(x,f) (44)
where we introduced the particle flux density J(x) which we can write by using (1221) in the form

J(x,t)= J -^ T {p( x + r,t)-p(x-r,t)}d"r, < S - (n - 1) < 2 (45)

which is determined up to an unimportant rotational gauge field (describeing closed flux lines which
do not change the local density fj43|) ). We observe that equal distributions p(x) = const would not
cause any particle flux. We can write the flux density also in the equivalent form

J(x, t) = (5 _ ( l l _ 1)) j J ^TP(x - r, *)d«r, < S - (n - 1) < 2 (46)

where a further equivalent expression is obtained by exchanging r — > — r in fj46|) . For n = 1 these
relations recover the expression found earlier [12]. We can conceive fT45]) or equivalently (SSJ) as the
(non-local) constitutive law connecting particle flux and density replacing the (local) Fick's law of
Gaussian diffusion.

Let us consider the particle jump rate into a small volume element SV around r = due to a
localized distribution Q(x, t = 0) = <5 n (x). This localized particle distribution induces at spacepoint
of distance x = |x| instantaneously the flux (due to non-local particle jumps of distance x)

]| ^" c »° [ t -(f-i)){ ^ < 47 >

These non-local particle jumps due to the flux (HTj) must cause at spacepoint x a particle balance
which must be positive due to counting particle jumps from r = to x, i.e. jumps over a distance x.
We obtain with (EFT

J^Q(x,t = 0) = -V x - J(x,t = 0) = jx-^ 1 >0, Vx^O (48)

which is positive and scaling as ~ x~ 5 ~ l and decays always with distance x whatever the dimension
n since n<S + l<n + 2 and is nonzero in the entire space whatever the jumping distance x of the
particles. The relation (j48p holds everywhere except in the origin x = 0. Due to the stochastic spatial
isotropy of the particle jumps, integration of fl4"8"l) over the unit sphere gives the jump rate of all jumps
with distance R at t = 0+. This rate is given by

10

n _ r

-/ J • nir -1 (K2(n) = -^rR n ~ 1 -- R~ s ~ l ~ IT 01 ' 1 , < a = 5- (n- 1) < 2, i? ^ (49)

where ^nj-R n_1 is the surface of the sphere of radius R. We can interprete (149]) as follows: The

probability that a particle which is at t = located in the origin r = undertakes a jump of distance
R within the infinitesimal time interval 5t scales as It follows that the jump rate of all jumps

of distance R at t = 0+ decays spatially scaling as ~ with the jump distance R and indeed is

Levi distributed where < a < 2 is the band of admissible Levi-parameter a whatever the dimension
n of the physical space.

A crucial rule plays the space-time representation of the time evolution operator from which we
considered the small time regime in (148]) . This propagator is generally defined by

Q(r,t) = e- tc ^5 n {r) (50)

where 5 n (r) denotes the n- dimensional Dirac's 5-function. In the stochastic picture the interpretation is
as follows: The kernel Q(r,t) describes then the conditional probability density and Q(r,t)d n r denotes
the probability to find the particle which was located at t = in the origin r = at time t in the volume
element d n r attached to the spacepoint r. Correspondingly Q(r,t) fulfills then the initial condition

QM = 0) = cT(r) (51)

and is hence itself a normalized probability distribution solving (134]) . We deduce some integral relations
thereof in the appendix |9j The density (139]) can then be represented by the convolution

p(r,t) = |Q(r-r',t)p (r')dV (52)

Taking into account that

together with the property

6 n (r) = —V / e ik ' r d n k (53)
(2tt)" '

£^ k ' r = u 2 n m s (ky k - r , m = 0, 1, 2, .. G N (54)

where u^ s (k) denotes the dispersion relation (123]) . we can write for any sufficiently smooth function
/(£) the relation f(tC nt s)e lk ' r = f(tu 2 s {k))e lk ' r . By using this property for the exponential operator
g-A^jt we obtain Q(r,t) in the form

g(r ' t} = (2^F J e ~ An ' skat ^' r ^ (55)

where < a = 5 — (n — 1) <2. The kernel is spatially isotropic and depends only on r = |r| due to
the isotropic symmetry of the 5-function. We note that the linear order in t of this Fourier integral
coincides with (J4BJ) constituting the regime of small times t (appendix [9]).

Distributions of the form (155]) are referred to as Levi distributions [HE]. In contrast to Gaussian
distributions, Levi-distributions exhibit diverging mean fluctuations (all even moments are diverging).
This can be directly verified from

< r 2 >= || Q(r,t)rV ik - r d"r| fc _ o = -V fc ■ V k Q(k,t)\ k=0 oo (56)

11

which is fulfilled by Q(k,t) = e~ An < skat for positive a in the interval < a < 2. It is interesting to see
that the condition of existence of the Laplacian (I19p leads to the same admissible a-band < a < 2 as
the condition of divergence of the variance (15*5]) . equivalent with the condition of non-differentiability
of Q(k,t) at k = 0.

All odd moments are vanishing due to the isotropic symmetry of the distribution Q(r,t). For the
further evaluation it is convenient to introduce the function defined by the surface integral over the
unit sphere

°- M = wf' L, inik)eTh = Wrl s{k ~ 1K " d " k (57)

where k\ denotes any Cartesian component of the unit vector k. We observe due to the spherical
symmetry that G n (r) contains only even powers in r, i.e. only the even part coshr/ci of the integrand
contributes to (1BT1) . The surface integral ( 15T|) is evaluated explicitly in appendix M (eq. (I109p ). The
kernel (155]) can be further written as

Q(r,t) = r- n P(^f) (58)

where P is a function of the "scaled time" X a = only. In the following we keep in mind that the
constant As jTl depends on n and 5 and skip these subscripts. The function P takes then the form

P(X)= e-^G^OC- 1 ^ (59)
Jo

To consider large A" = ^ » 1 we can write by introducing the "fast" variable u = A£, and we
emphasize that < a < 2 and hence - is a positive exponent, so we can rewrite

POO

P(A) = A~ n / e- ua u n - l G n {i\- l u)du (60)
Jo

and hence the kernel Q = r~ n P has the form

Q ^ = X^ W ((£))' 0<« = *-(n-l)<2 (61)

W(\~ a ) = r e- ua u n - l G n {i\' l u)d Ul A = (62)

o

and the function G n depends only on the dimension n of physical space and is defined by (1571) . We note
that equations (|6T]) with ( 162]) are exact relations. Generally, except in certain cases to be considered,
the integral (I62p cannot be obtained in closed form. Despite a = 2 is a forbidden case in our model,
we can formally consider this case for which ( I6"T|) with (I6"2"|) can be evaluated in closed form and lead
to a Gaussian distribution

1

Q ^ t) = ^Mji e 4M (63)

where here A = A 2 which is not defined by our model as dispersion relation (12~"I|) is diverging for
a = 2. Representation ([61]) is in a sense analogue to the Gaussian distribution (1631) where the time
dependence of the normalization factor of (I6T1) is given by (At)~& and exhibits in the Gaussian case
(a = 2) (At)~2 and leading to

12

Asymptotic regime \ a = >> 1 where A" denotes a scaled time: In this regime, relation
fl62|) assumes asymptotically the form

G (0) ( r 2
Q{r, At » r a ) ~ I(n, a) - —rl{n + 2, a) + ..0(

(At)* V (At);

(64)

G w (0)(/(n,q)
^ g(t) = ^Ap ^°

which decays in the dominant term in time as t~& independent on r and where

I(n,a) = / e^V-Mu (65)

JO

and

G - (0) = wkw) (66)

The asymptotic relation fl64|) describes the manner how the distribution approaches the thermody-
namic Boltzmannian equal distribution. Since f l64"j) does not contain any information about the initial
positions of the particles we can conclude that the leading term Q(t) in flMj) is universal and holds
for any initial distribution po(r). This follows also from the asymptotic relation where we assume
that At/R a » 1 and R » 1: The leading contribution to the density in this regime then is (with
Q(r,t)~Q(t))

p(r, t) = J Q(r - r', t)p (r')dV ~ Q(t) | p (r')dV = Q(t) (67)

which leads indeed for any initial distribution po(r) to the identical spatially uniform asymptotics
Q(t) ~ t~~ — > where the exponent depends only on the ratio - of the physical dimension n and
the Levi parameter a and with the restriction | < - < oo. If t exceeds all limits the system of
diffusing particles "forgets" its past and approaching all the same thermodynamic attractor Q(t) — > 0.
However, since the scaled time At/r a is for finite t never a large quantity in the entire space as there is
always a region where At/r a << 1 is still small and hence ( l64"j) is not (yet) valid there. In other words:
The spherical region for which the spatially quasi-equal distribution Q(t) of ( I64p is valid and hence

which is close to thermodynamic equilibrium is expanding in time slower than r(t) ~ (At)*. The
volume of this expanding region scales as V(t) ~ r n (t) ~ (At) a. The assymptotic uniform distribution
Q(t) ~ 1/V(t) ~ (At)~~ scales as the inverse of the volume V(t) of thermodynamic quasi-equilibrium
where Q(t) is the quasi-equal distribution within this region of "quasi-equilibrium". The universal
scaling behavior of the type fl64"|) in fractional models of anomalous diffusion was already noted earlier

HZ].

For all dimensions n of the physical space holds: The algebraic decay of the kernel Q(r,t) ~ Q(t)
being independent from r for large scaled times At/r a » 1 is a necessary consequence of Boltzmann's
postulate saying that the location of a diffusing particle becomes completely undetermined without
any preferred location in the thermodynamic equilibrium indicating the complete "loss of information"
about the location of the particle. This corresponds to the thermodynamic necessary condition of
approaching maximum (infinite) entropy. We will demonstrate this by the following brief consideration
of the entropy which we define here as negative H-function [H]

S(t) ~ - J Q(r, t) log Q(r, t) d"r (68)
We are especially interested in this quantity in the large times limit t — > oo. In this regime we can put

13

S(t) ~ -log{(At)~«} [ Q(r, t)d n r ~ -log(t) oo (69)

which diverges logarithmically and spatially homogenously in time indicating in the stochastic picture
that the state of complete uncertainty about the location of the particle is approached as the global
time t tends to infinity It follows that (|69|) is universal and independent from the initial distribution
Pq(t). In other words relation (JBSJ) expresses the validity of the H-theorem due to Boltzmann which
is equivalent to the fact that distributions are broadening in time approaching equal distribution as
time tends towards infinity. The divergence of the maximum entropy in (1691) is reflecting the fact that
the available volume for any particle becomes infinite as t — > oo whereas the particle number remains
a finite constant. The entropy would approach a finite equilibrium value if for t — > oo the volume
acessible for the particles would be finite. A finite volume, however would be in contradiction to the
self-similarity and as a consequence non-locality of the Laplacian. For a "real" physical system in
nature therefore, self-similarity can only be approximatively fulfilled.

Let us briefly consider the one-dimensional case n — 1: There we have a = 5 — (n — I) — 5 with

Q 1 (r, At » r s ) ~ ^^-(At)^ , < 5 < 2 (70)

7T

and

1(1,5) = / e~ u du (71)

Expressions (170)) with ( 171 j) is in accordance with the asymptotic relation obtained in [12] for the
one-dimensional case. There is one single case remaining, where Q(r, t) can be obtained in closed form:
From ( )62|) that this is the case for a = n which can be only fulfilled for n < 2. Since n G IU the only
case is n = 5 — 1. We obtain then

and (165]) yields

Gi(iO = -cos£ (72)

7T

jy (A" 1 ) = - [e- u cos (A- 1 ?/) dw = - (^ 2 9 (73)

The distribution ( IBTj) is then obtained as (r = |x|)

1 /If

« r - ( ) = ^(S)^ (74)

which is known as Cauchy distribution [2] which has also the property of diverging even moments.
The Cauchy distribution hence is the result of a Levi flight for a special Levi parameter a = 5 = 1.
Sometimes this motion is referrred to as Cauchy flight. The Cauchy case stands out in the present
model and appears uniquely for n — 5 — 1. The Cauchy distribution behaves for large At » \x\ as
Q(t) ~ l{At)~ l in accordance with flMJ) for a = n = 1 (with 1(1, 1) = 1 and Gi(0) = J).

6 Conclusions

We have deduced a Laplacian operator for a n-dimensional infinite space with self-similar symmetry.
The Laplacian introduced in this paper fulfils all necessary requirements: linearity, self-adjointness,
isotropic symmetry, negative semi-definiteness, and translational symmetry. The self-similar symmetry
makes this Laplacian operator non-local. To have a physical picture, the self- similarity can be conceived

14

as an elastic medium with self-similar scaling invariant harmonic interparticle interactions. Employing
this picture we have deduced the dispersion relation with the density of normal oscillator modes. The
density of normal modes in the frequency space fulfills a scaling law ~ cu^^ 1 with a characterstic
strictly positive exponent being always greater than n — 1 which would be the exponent due to a
traditional Laplacian of the n-dimensional space.

We analyzed a diffusion problem defined by a Fokker-Planck equation by employing the self-similar
Laplacian. The model describes anomalous diffusion allowing non-local particle jumps where the jump
distances are Levi-distributed (Levi flights). The solutions of the Fokker Planck equation are Levi-
stable distributions with characteristic algebraic decay ~ in the regime of large scaled times with
the Levi parameter a being in the interval < a < 2 whatever the dimension n. The present model of
anomalous diffusion could have some applications on physical phenomena dominated by non-Brownian
erratic motions such as for instance in turbulence. We hope the present approach inspires further work
in such directions.

Moreover, the present approach can be used as a starting point to describe continuous field problems
governed by self-similar constitutive laws in statics and dynamics. The physical nature of the fields
can be of any kind, such as of mechanical or electromagnetical nature. A corresponding self-similar
electromagnetic field theory will be presented in a sequel.

7 Acknowledgement

This paper is dedicated to the memory of our dear friend Prof. Arne Wunderlin (Stuttgart, Germany).

8 Appendix

8.1 Derivation of the vector field (122

Here we deduce the vector field ("dielectric displacement") D of relation ( 12"2"|) . To this end we start
with the Laplacian f JT9|)

h s I (w(x + r) + w(x — r) — 2w(x))

A M)M (x) = £ / WAT ^ T X ; < S - (n - 1) < 2 (75)

and use the identity

r-'-^J^^^-ir-'- 1 r) (76)

with V • (a&) = 6V ■ a + a ■ Vb. We note that < 5 — (n — 1) < 2 is always non-zero. By applying the
Gauss theorem for the boundary integral term we get

J r-- & - 1 (u(x + r) + u(x - r) - 2w(x)) d n r =

lim/ r^dfifa) • | -= — p—-r ■ (r~ s ) (u(x + r) +u(x-r) -2u(x))l+ (77)
r-^coJqv(r) [d — (n — l) J v '

1 r

+ - — -. — / (r-^r) ■ V r (u(x + r) + u(x - r) - 2w(x)) d n r

o — (n — 1) J

The boundary integral over dyim scales as i?~( <5 ~( n_1 )) — y (0 < 5 — (n — 1) < 2) and is hence
vanishing as R — > oo (5 — (n — 1) > 0). When we further use V r (w(x + r) + tt(x — r) — 2u(x)) =
V x (w(x + r) — w(x — r)) we can write (|75p as a divergence

15

A M) u(x) = V x D (76
with the vector field D being determined (up to a unimportant rotational field)

1 h 5

D(x) = - - — /V^r {u(x + r) - u(x - r)} d n r , 0<<J-(n-l)<2 (79)

o — (n — 1) 2£ J

which is relation ( 122]) . We observe that the integrand is an even function of r. Taking into account
that the volume integral over any odd function of r is vanishing we can also write

D(x) = ^77jbl) f / ^ x + ^ ' < * - (n - 1) < 2 (80)
or equivalently (by replacing x + r — > r)

D(x) = - I - g / -^nfrjd-r , < 5 - (n - 1) < 2 (81)

5 - (n - 1) C

where only the even part of the integrand namely f !79p ) contributes. All representations (I79p and

or flHTj) exist in the band < 5 — (n — 1) < 2 just as the Laplacian (j75]l . flHUl) can be further written
in the form of a gradient of a scalar potential

$(x)

— /'|x-r| 1 - 5 u(r)d n r 5^1 0<<J-(n-l)<2 (83)

1 J C J

D(x) = V x $(x) (82)
where the scalar potential $(x) can be written as

(*-(n-l))(*-l)C

For n — 1 this expression recovers those obtained in our previous paper ([9]). We note that the
scalar $(x) is a convolution of the scalar field u with the convolution kernel being the power function
("Riesz potential") |r — x| 1_<5 where n — 1 < 5 < n+ 1. The special case of 5 = 1 is appears only for
n = l. For n > 2 we have l<n — l<5<n + l and hence 5 > 1 for dimensions n = 2, 3, ... In
these cases i? 1-5 is vanishing at infinity and singular at R = 0. Only the case of one dimension n = 1
with < 5 < 2 has for < 5 < 1 the "anomaly" that the Riesz potential is diverging at infinity and
vanishing in the origin. It follows that

A n , 5M (x) = V x • V x $(x) (84)

where V x • V x denotes the traditional Laplacian of the n-dimensional space. Relation f[M|) recovers
for n = 1 the expression we obtained earlier [9].

The case 5 = 1 which can only occur for n = 1 furthermore again stands out among all the others.
This case e.g. lead us to the Cauchy-distribution in section |5j Evaluation of ( |83|) for n = 5 = 1 yields

h f°°

$ 5 =i,„=i(x) = -- / lnflz - r|)«(r)dr (85)
which has been already obtained in [9j.

16

8.2 Determination of A n j

In this appendix we determine the constant A n j = u 2 s (k = 1) occuring in the dispersion relation fl25|)
u 2 s (k) = A n ^k a (a = S— (n— 1)). We introduce r ■ k = krrii, (n • k = krii). Then we have to evaluate

(El

u 2 (k = l) ni5 = A niS = -J a A 1;0l (86)

with

= "Us= a (k = 1) = ^ f (1 Z S i (T)) dT 0<a = 5-(n-l)<2 (87)

C A,

and

J a = dn(n)\ ni \ a 0<a = 5-(n-l)<2 (88)

J\n\=l

In view of (I86p we can conclude that the dispersion relation of the n-dimensional space corresponds
to that one in one dimension when we replace 5 — > a up to a prefactor 4f which depends only on
the exponent a and dimension n of the physical space. The constant A\^ a of (IH7|) is nothing but
the dispersion relation for the one- dimensional case u 2 =is (k = 1) which we already deduced in the
one-dimensional model [12]. For the evaluation we refer to that paper. We obtained there

A ha = - ^ wa > 0, 0<a = 5-(n-l)<2 (89)

Qa\ sin 1 y

where this constant is strictly positive and finite in the admitted a-range. For physical dimensions
n — 1,3 fl86|) can be explicitly evaluated:

Case n = 1: Then a = 5 and J$ = 2 and so (186]) yields then our previously deduced expression of [12]

C<5! sin f

Case n = 3: a = 5 — 2 and the integral fl88|) yields then

A M = T7i^7 >0, 0<5<2 (90)

47T

^=5-2 = (91)

o — l

and hence

2h 5 7c 2

A ^= -7Z . „ ( ,_ 2) >0, 2<<5<4 (92)

C(o - 1)! sin v 2 ;

In all cases the necessary condition oo 2 (l) > which follows already from the positiveness of the
integrand of (18"T|) .

9 Some useful integrals

In this appendix we deduce a rather remarkable relation when we take into account the following
identities, namely those of equations (jlSjl and (1501 being the linear order in t of relation ( 155|) . First
we have

f (r, t = 0) = A^-M = - At / e'-rd»k = £ / ^A_£O d V = £-«- (93)

17

where the latter intergral is obtained by application of the Laplcian on the 5-function for r ^ and
yields ( I48p . The Fourier integral can be rewritten for r ^0

f M = °> = -^f G " (iT)TSdT (94)

where G n is the surface integral defined in (|57p which is evaluated in closed form below. Comparision
with the explicit expression ( I48p yields the integral relation

/ G n (ir)t S dr = -—— < (95)
which can be exlicitely confirmed by evaluating the left hand side for n = 1 and n = 3.
Surface integrals

We choose a Cartesian coordinate system with and n = (ni, .., n n ) tr with n ■ n = 1 being a
parametrization of the n-dimensional unit-sphere. Let us then evaluate the integral of an integer
power of any Cartesian coordinate

A£> = I dft(n)n™, m = 0,1,2,.. e Mo (96)

J\n\=l

where we integrate over the surface of the n-dimensional unit-sphere. The superscript (■■y^ indicates
the dimension (n = 1,2,3,. G NJ) of the space. We observe that all integrals of odd powers (odd

functions) -4.2m+i = are vanishing. Only integrals over even powers are non-vanishing and by
applying the Gauss-theorem we find the recursion

(n) _ (2m- 1) (n)
{2m — 2 + n)

where n denotes the dimension of the space n = 1,2,3,... This recursion can be applied m-times
linking A^L with A^ = O n (l) which is the surface of the unit sphere. In this way we obtain

A (») - (2m-l).. x3x 1 (n)

^ 2m "(2m-2 + n)..x(n + 2)x n [ }

The products (each containing m factors) in the nominator and the denominator can be written as

(2m-l)..x3xl = M (99)

2 m ml

and

So we obtain for (198

(o. _ i _l m ))\

(2m-2 + ra).. x (n + 2) x n = 2 m K2 (100)
(n) _ (2m)!(f-l)!

where O n (l) denotes the surface area of the unit-sphere embedded into the n-dimensional space

2ii2 2tt2

= (Fiji = FM (102 >

18

and hence

(n) 27rf(2m)! _ 2^(2m)\

2m 2 2m m!(f -1 + m)! 2 2m m!r(§ + m) 1 J

So we can perform the unit-sphere surface integral of any sufficiently smooth function /(r)

/(W = E (104)

g{i) = [ dfi(n)/ (£m) (105)

■/|n|=l

where only its even part tilhJizll (even powers) contributes. By using (j96p - (ll03j) we obtain for (11051)
the series

00 An)

*®=]Z<*>^f (106)
which is uniquely determined by all even derivatives 0-2™ of f(r) at r = 0.
Explicit evaluation of ( I57|)

With ( I106p we can evaluate the function G n {i^) which is defined by surface integral (J57j) . The inte-
grand (I104p is /(£ni) = (2~^ cos £ n i ( or alternatively /(£ni) = ^ys" e ^ ni s i nce an Y °dd function such
as sin£ni yields a zero contribution). Then we obtain for G n (i£) with (I106p the following series

GM) ^E ( - ir ^w^ (107)

Taking into account the generalized definition of the Bessel function of the first kind Q16J p. 360,
9.1.10.) for any (integer and non-integer) v G fR

j -H91 U(™Ui) (^T « io8)

So we obtain for (|5T|) the closed form expression

G„(ifl = 7^7^1^-1(0 (109)
(2tt)2^ 2

which holds for any dimensions n G IU and is even valid for positive non-integer dimensions since (llOip
with (I102p assume the same form when they are deduced for non-integer n > 0. Let us now look
especially on the physically important cases of dimensions n — 1, 2, 3.

(i) n = 1

We have with

2 2rn m\(m - -)! = (2m) (2m - 2). .2 x (2m - l)(2m - 3)..3T(-) = {2m)\^ (110)

and hence so

19

so that

G n= i(iO = -cos£ (112)

7T

which we reverify to be correct by its definition G n= i(£) = ^ (e^ + e -1 ^.

(ii) n = 2

Then we have

J (f) = V (-l) m — i (113)
^ 2 2m m!m! v ;

and

Gn= 2 (^) = (^J-MO (H4)

which we can again reverify directly by the definition of G 2

G ^ )= w?L s ™" A,p (115)

by taking into account the definition of Jo-

(iii) n = 3

Finally we have

2 2m m!(m+^)! = (2m+l)(2m-l) x ..3 ! x 2m(2m-2) x ..2 = (2m+l)! Q^! = (2m+l)!^ (116)
as = |r(|) = ^ and hence

/ ot oo t2m / o

J ^ = lTE, { - 1)m d+w = h^ (117)

so that we obtain

G„ 3 «) = 5^ (118)
which again is directly verifiable by the definition

References

[1] B. Mandelbrot, The Fractal Geometry of Nature, B. Mandelbrot, W. Freeman & Co 1982, ISBN
0-716-71186-9).

[2] Mandelbrot B., Fractales, hasard et finance, B. Mandelbrot 1997, ISBN 978-2-0812-2510-7 (pages
106ff, 156).

[3] Han Qing, Lin Fanghua (1997), Elliptic Partial Differential Equations, New York, Courant Insti-
tute of Mathematical Sciences, ISBN 0-9658703-0-8.

20

[4] P. Levi, Processus stochastiques et mouvement Brownien, Reprint, Editions Jacques Gabay, 1965.

[5] Glockle W. G. and Nonnenmacher Th. F., A fractional calculus Approach to Self Similar Protein
Dynamics, Biophysical Joural, Vol. 68, (1995), pp. 46-53.

[6] HG. Sun, W. Chen, C. Li, YQ. Chen, Fractional differential models for anomalous diffusion,
Physica A 389 (2010) pp. 2719-2724.

[7] F.Lenz,T.C.Ings, L. Chittka, A.V. Chechkin, R. Klages, Spatiotemporal Dynamics of Bumblebees
Foraging under Predation Risk, PRL 108, 098103 (2012).

[8] Lifshitz, E. M.; Pitaevskii, L. P. (1981). Physical Kinetics. Course of Theoretical Physics. 10 (3rd
ed.). Pergamon. ISBN 0-08-026480-8.

[9] T. M. Michelitsch, G. A. Maugin, F. C. G. A. Nicolleau, A. F. Nowakowski, and S. Derogar,
Dispersion relations and wave operators in self-similar quasicontinuous linear chains, Phys. Rev.
E 80, 011135 (2009).

[10] T.M. Michelitsch, G.A. Maugin, F.C.G.A. Nicolleau, S. Derogar, Wave Propagation in Quasi-
continuous Linear Chains with Self-similar Harmonic Interactions: Towards a Fractal Mechnics, in
Mechanics of Generalized Continua, Advanced Structured Materials, H. Altenbach, G.A. Maugin,
V. Erofeev eds., Springer Berlin, Heidelberg, Vol. 7 (2011), pp. 231-244. Online ISBN 978-3-642-
19219-7.

[11] T. M. Michelitsch (2011), The self-similar field and its application to a diffusion problem, J. Phys.
A: Math. Theor. 44 465206.

[12] T. M. Michelitsch, G. A. Maugin, M. Rahman, F. C. G. A. Nicolleau, A. F. Nowakowski, and S.
Derogar (2012), A continuum theory for one-dimensional self-similar elasticity and applications
to wave propagation and diffusion, European Journal of Applied Mathematics, in press.

[13] Peitgen H.-O., Jurgens H., Saupe D., Fractals for the Classroom: Part 1: Introduction to Fractals
and Chaos, Springer (1991), ISBN-10: 9780387970417.

[14] V.E. Tarasov, Continuous limit of discrete systems with long range interaction, J. Phys. A: Math.
Gen. 39 (2006), 14895-14910.

[15] A.C. Eringen, Nonlocal Continuum Field Theories, 2001, Springer New York, Berlin, Heidelberg,
ISBN 0-387-95275-6.

[16] M. Abramowitz, LA. Stegun, eds. Handbook of Mathematical Functions with Formulas, Graphs,
and Mathematical Tables. New York: Dover, 1972. (chapter 6).

[17] R. Metzler, J. Klafter (2000), The random walk's guide to anomalous diffusion: a fractional
dynamics approach, Physics Reports 339/1, pp. 1-77.

[18] H. Goldstein (1980), Classical Mechanics, 2nd ed. Addition Wesley, pp. 35-69.

21